Cray Research Inc., a computer maker better known for supplying research scientists than data analysts, would nevertheless like to be your analytics vendor. The company is releasing today the Urika-GX Agile Analytics Platform, a supercomputer for the rest of us that fits in a standard data center footprint and supports a host of open source big data options, as well as some of its own secret recipes.

The 44-year-old supercomputer company, which was founded by brilliant-but-reclusive hardware engineer Seymour Cray, says there’s good reason to take it seriously as an analytics player. “We’ve been in the analytics space for a long time,” said Ryan Waite, senior vice president of products. “We were selling Cray systems to Wall Street for use in analytics in the 1980s.”

In fact, the Urika-GX is unlikely to be used for garden-variety business intelligence. With support for up to 1,728 cores per system, up to 22 terabytes of memory and 227 terabytes of combined SSD and disk storage, the system is optimized for very large problem-solving, such as finding the strongest relationship clusters in graphs containing billions of relationships. It comes with a full Hadoop stack installed, including Apache Mesos for resource management, Yarn for cluster management, Spark analytics and the HDFS file system.

There’s also a proprietary Cray Graph Engine built on top of a native implementation of the Simple Linux Utility for Resource Management (SLURM) job-scheduler. Graph engines are used to rapidly define and test relationships between nodes in network. One popular application of graph computing is recommendation engines.

The Urika-GX uses OpenStack for management and supports many popular analytics tools such as Kafka, Cassandra, Lustre and R. The analytics-in-a-box packaging is intended to address the complexity IT organizations confront as they piece together analytics solutions into different configurations that Waite called “franken-clusters.”

All-in-one

“We wanted to build a platform that could run multiple analytics workloads in the same box and concurrently,” he said. By using Mesos for resource management, “one day you can have half the cluster supporting Spark and the next day only one quarter.”

Of course, the system isn’t entirely off-the-shelf. It uses Cray Aries fabric in the backplane and a Cray version of the CentOS operating system, as well as Cray’s optional Sonexion storage, which is optimized for transferring very large files from storage to memory. All of this adds up to performance that Cray claims is nearly double that of conventional analytic systems for graph loading and four times faster as measured by the PageRank benchmark. Enhancements to the graph engine have improved performance by about 20-fold compared to previous versions.

Cray is targeting the machine at compute-intensive vertical markets like cybersecurity, financial services, life sciences, weather forecasting and engineering, but it isn’t ruling out any market that needs to process large amounts of data involving complex relationships.

Waite said the popular perception that supercomputers are only for massive corporations and government research centers is exaggerated. “People think supercomputers are expensive, but the fact is that supercomputing is often price-competitive,” he said. While declining to specify the Urika-GX’s price, he said it will be competitive with that of other makers of specialized analytics hardware.